WO2020235081A1

WO2020235081A1 - Quantization device, quantization method and program

Info

Publication number: WO2020235081A1
Application number: PCT/JP2019/020468
Authority: WO
Inventors: 小軍ウ; 正樹北原; 清水　淳
Original assignee: 日本電信電話株式会社
Priority date: 2019-05-23
Filing date: 2019-05-23
Publication date: 2020-11-26
Also published as: JP7284429B2; US20220222860A1; JPWO2020235081A1

Abstract

A quantization device according to the invention is provided with: an imaging unit that converts statistical data at real-space positions to the pixel values of coordinates associated with the positions in an image; and a deriving unit that derives a quantization parameter corresponding to the quantization width of a pixel value for each of one or more real-space positions with respect to a part of or the whole of the image. The quantization device may be further provided with a target acquisition unit that acquires information representative of a target related to statistics. The deriving unit may derive a quantization parameter corresponding to a quantization width that is longer, the greater the pixel value is.

Description

Quantizer, quantization method and program

The present invention relates to a quantization device, a quantization method and a program.

There is a demand to extract semantic data (semantic data) from a huge amount of statistical data. As an image used to satisfy this demand, an image representing statistical data (hereinafter referred to as "statistical image") is generated (see Patent Document 1). The coordinates in the statistical image are associated with the positions in real space. The pixel value of a pixel in a statistical image represents statistical data at a position in real space associated with the coordinates of that pixel.

For the purpose of compressing the capacity of such a huge amount of statistical data, the pixel value of each pixel of the statistical image is based on an image coding standard such as HEVC (High Efficiency Video Coding) standard (see Non-Patent Document 1). May be quantized.

JP-A-2018-128731

The pixel value of the pixel of the statistical image represents the statistical data. Therefore, the quality of the image quality of the statistical image is not evaluated by humans. However, the conventional quantization method of the image coding standard is a quantization method that emphasizes improvement of image quality. In order to emphasize the improvement of image quality, in the conventional image coding standard, the same quantization parameter (QP: Quantization Parameter) is used for all pixels in the statistical image when the pixel value is quantized. The quantization parameter is a parameter used to determine the quantization width of the pixel value.

When the same quantization parameter is used for all pixels in the statistical image, the statistical properties cannot be maintained in the statistical image due to changes in the ratio (dense and dense relationship) between the statistical data represented by each pixel value. There is. When the statistical data is, for example, a population, the influence of the quantization parameter on the pixel value differs greatly between the pixel value representing a large population in the city center and the pixel value representing a small population in the mountainous area. For example, if the quantization parameter is set according to the pixel value representing a large population in the city center, each pixel value representing a small population in each mountainous area will be averaged, so that it represents a small population in each mountainous area. Statistical differences between pixel values are lost. In this way, it may not be possible to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties.

In view of the above circumstances, it is an object of the present invention to provide a quantization device, a quantization method, and a program capable of quantizing the pixel values of the pixels of a statistical image so as to maintain statistical properties. ..

One aspect of the present invention is an imaging unit that converts statistical data at a position in real space into a pixel value of coordinates associated with the position in an image, and a quantization parameter corresponding to the quantization width of the pixel value. Is a quantization device including a derivation unit for deriving one or more positions in the real space for a part or the whole of the image.

According to the present invention, it is possible to quantize the pixel values of the pixels of a statistical image so as to maintain the statistical properties.

It is a figure which shows the structural example of the quantization apparatus in 1st Embodiment. It is a figure which shows the example of the statistical image in 1st Embodiment. It is a figure which shows the example of the prediction image in 1st Embodiment. It is a figure which shows the example of the difference image in 1st Embodiment. It is a figure which shows the example of the pixel value of the predetermined pixel of the difference image in 1st Embodiment. It is a figure which shows the example of each pixel value of the predetermined pixel group of the difference image in 1st Embodiment. It is a figure which shows the example of the differential decoding image in 1st Embodiment. It is a figure which shows the example of the decoded image in 1st Embodiment. It is a flowchart which shows the operation example of the quantization apparatus in 1st Embodiment. It is a figure which shows the structural example of the quantization apparatus in 3rd Embodiment. It is a figure which shows the example of the statistical image in 3rd Embodiment. It is a figure which shows the example of the prediction image in 3rd Embodiment. It is a figure which shows the example of the difference image in 3rd Embodiment. It is a figure which shows the example of each pixel value of the predetermined pixel group of the difference image in 3rd Embodiment. It is a figure which shows the example of the image which shows the difference data in 3rd Embodiment. It is a figure which shows the example of the decoded image in 3rd Embodiment. It is a flowchart which shows the operation example of the quantization apparatus in 3rd Embodiment.

Embodiments of the present invention will be described in detail with reference to the drawings.
(First Embodiment)
FIG. 1 is a diagram showing a configuration example of the quantization device 1a. The quantization device 1a is a device that quantizes the pixel value of each pixel of the statistical image. The coordinates in the statistical image are associated with the positions in real space. The pixel value of a pixel in a statistical image represents statistical data at a position in real space associated with the coordinates of that pixel. The statistical data may be, for example, data representing the population or data representing the demand for taxis (for example, the number of dispatched vehicles, the number of candidate passengers). Since the pixel value may not fit in the permissible range, the statistical data may be converted to a value within the permissible range by executing a predetermined conversion process on the statistical data. Good. As the predetermined conversion process, for example, the conversion process disclosed in Reference Document (Japanese Unexamined Patent Publication No. 2017-123021) may be used.

The quantization device 1a includes a coding device 2a (quantization unit) and a decoding device 3 (inverse quantization unit). The coding device 2a includes an imaging unit 20, a derivation unit 21, a subtraction unit 22, and a difference quantization unit 23. The decoding device 3 includes an inverse mapping processing unit 30, a difference decoding unit 31, and an addition unit 32.

Part or all of the quantization device 1a is software in which a processor such as a CPU (Central Processing Unit) executes a program stored in a memory which is a non-volatile recording medium (non-temporary recording medium). Is realized as. The program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, flexible disks, magneto-optical disks, portable media such as ROM (ReadOnlyMemory) and CD-ROM (CompactDiscReadOnlyMemory), and storage of hard disks built in computer systems. It is a non-temporary storage medium such as a device. The program may be transmitted over a telecommunication line. A part or all of the quantization device 1a is, for example, an electronic circuit (Field Programmable Gate Array) using an LSI (Large Scale Integration circuit), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), an FPGA (Field Programmable Gate Array), or the like. It may be realized by using hardware including electronic circuit or circuitry).

The coding device 2a is a device that quantizes a statistical image expressed in an uncompressed image format (hereinafter referred to as "uncompressed statistical image") and encodes the quantized statistical image. The imaging unit 20 acquires statistical data associated with a position in real space. The imaging unit 20 generates an uncompressed statistical image based on the statistical data. The imaging unit 20 converts the statistical data at the position in the real space into the pixel value of the pixel of the coordinates i (x, y) associated with the position in the real space in the image. For example, the imaging unit 20 may generate a statistical image by using the data processing method described in the reference (Japanese Unexamined Patent Publication No. 2017-123021). The imaging unit 20 outputs an uncompressed statistical image to the derivation unit 21 and the subtraction unit 22.

The derivation unit 21 acquires an uncompressed statistical image from the imaging unit 20. The derivation unit 21 executes prediction processing (for example, intra-prediction, time-direction prediction) on the uncompressed statistical image.

Deriving section 21, a quantization parameter corresponding to the quantization width d _i of the pixel values of the uncompressed statistical image is derived for each pixel of the coordinate i in part or all of the statistical image. The longer the quantization width d _i, the value of the quantization parameter is large. Deriving unit 21 derives the quantization parameter corresponding to the long quantization width d _i greater the pixel value V _i. Deriving unit 21, the pixel value V _i of the coordinate i of the statistical image, executes the quantization processing by the quantization width d _i determined according to the quantization parameter determined for each pixel. Quantization width _{d i} is expressed by the equation (1).

_{_{d i = f (V i)}} ... (1)

Here, i is an index and represents the coordinates in the statistical image corresponding to the position in the real space. V _i represents the pixel value of the coordinate i. f represents a predetermined proportional function.

The derivation unit 21 outputs the quantization parameter determined for each pixel to the difference quantization unit 23. Deriving unit 21, the quantization width d _i instead of the quantization parameter may be outputted to the difference quantization unit 23.

Deriving unit 21, the data (hereinafter referred to as "feature data".) The resulting image feature amount of the statistical image against prediction processing including quantized pixel value in the quantization width d _i of each pixel, Output to the reverse mapping processing unit 30. The derivation unit 21 maps the feature amount data (low-dimensional data) when the high-dimensional data can be mapped to the low-dimensional data by a predetermined function and the reverse mapping of the mapping result can also be calculated. The resulting data. Inverse mapping processing (processing for returning feature data to high-dimensional data) is executed for the feature data (low-dimensional data) obtained as a result of the prediction processing. The derivation unit 21 outputs the prediction image (prediction signal) obtained as a result of the inverse mapping process to the subtraction unit 22.

The subtraction unit 22 acquires an uncompressed statistical image from the imaging unit 20. The subtraction unit 22 acquires the predicted image obtained as a result of the reverse mapping process from the derivation unit 21. The subtraction unit 22 subtracts the pixel value of the predicted image from the pixel value of the uncompressed statistical image for each pixel. That is, the subtraction unit 22, a pixel value which are not quantized by the quantization width d _i and (non-pixel value of the compressed statistics image), the quantized pixel value in the quantization width d _i (pixel value of the prediction image ) And the error value (residual) are derived for each pixel. The subtraction unit 22 outputs the subtraction result to the difference quantization unit 23 as a difference image.

The difference quantization unit 23 (compression processing unit) acquires a difference image from the subtraction unit 22. The difference quantization unit 23 acquires the quantization parameter from the derivation unit 21. Difference quantization unit 23, the pixel values of the difference image (error value) is quantized by the quantization width d _i determined according to the quantization parameter determined for each pixel.

Difference quantization unit 23, the error value of the predetermined criteria (e.g., 0) pixel having an error value in the range from the quantization width "-d _i" to the quantization width "+ d _i" around a, Detect in the difference image. That is, the difference quantization unit 23, the absolute value of the pixel values of the difference image (error value) is the equal to or less than the absolute value of the pixel of the quantization width "± d _i" is detected in the difference image. The difference quantization unit 23 quantizes the error value of the pixel detected in the difference image to 0. The difference quantization unit 23 outputs the pixel value of the difference image (hereinafter referred to as “difference data”) to the difference decoding unit 31.

The difference quantization unit 23 leaves the error value of the pixel not detected in the difference image as it is. Incidentally, the difference quantization unit 23, an error value V _i in the range of the quantization width "+ d _i" to the quantization width "D> (+ d _i)" may be replaced with the error value "+ d". By repeating the replacement process by the difference quantization unit 23, the continuous error value may be replaced with a plurality of discrete error values.

The decoding device 3 is a device that decodes a compressed statistical image based on the encoded statistical image. The inverse mapping processing unit 30 acquires feature data from the derivation unit 21. The inverse mapping processing unit 30 executes the inverse mapping processing on the feature data. That is, the inverse mapping processing unit 30 generates a predicted image based on the feature amount data. The reverse mapping processing unit 30 outputs the predicted image obtained as a result of the reverse mapping processing to the adding unit 32.

The difference decoding unit 31 acquires the difference data from the difference quantization unit 23. The difference decoding unit 31 executes decoding processing such as inverse quantization on the difference data. The difference decoding unit 31 outputs the result of the difference data decoding process (hereinafter referred to as “difference decoding image”) to the addition unit 32.

The addition unit 32 acquires the feature amount data from the inverse mapping processing unit 30. The addition unit 32 acquires the difference decoding image from the difference decoding unit 31. The addition unit 32 adds the pixel value of the feature amount data and the pixel value of the difference-decoded image for each pixel. The addition unit 32 derives the addition result as a statistical image (decoded image) expressed in a compressed image format. The addition unit 32 outputs the addition result to a predetermined external device.

Next, an example of quantization processing will be described.
FIG. 2 is a diagram showing an example of a statistical image. In the following, the statistical image is composed of "8 × 8" pixel groups as an example. In FIG. 2, the statistical image includes pixels 100-108. Hereinafter, the values described in the pixels in the figure represent the pixel values. For example, "156" described in pixel 100 in FIG. 2 represents the pixel value of pixel 100.

The pixel value of pixel 100 represents, for example, the population of a first area that is not a depopulated area (for example, an urban area). The pixel value of the pixel 100 at the coordinates (1,7) is "156" as an example. The pixel value of the pixel 101 represents, for example, the population of a second area that is a depopulated area (for example, a mountainous area). The pixel value of the pixel 101 of the coordinates (7, 1) is "6" as an example. Each pixel value of pixels 102-108 is "6" as an example.

Since the pixel value of the pixel 101 is smaller than the pixel value of the pixel 100, the quantization error allowed for the pixel value of the pixel 101 is smaller than the quantization error allowed for the pixel value of the pixel 100. If, for example, the same quantization width "10" is used for the quantization of the pixel value in all the pixels in the statistical image, the pixel value "6 (pixel value less than the quantization width" 10 ") of the pixel 101). "Is changed to" 0 ", and the statistical properties of depopulated areas may not be maintained. In addition, the statistical properties of the statistical image may not be maintained due to changes in the ratio (dense and dense relationship) between the statistical data (values) represented by each pixel value.

The derivation unit 21 derives a quantization parameter for each pixel value in order to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties of the depopulated area. For example, the derivation unit 21 derives a quantization parameter corresponding to the long quantization width d _i greater the pixel value V _i.

In FIG. 2, as an example, the derivation unit 21 derives a quantization parameter corresponding to a quantization width “1” less than the threshold value for a pixel having a pixel value less than “50”. As an example, the derivation unit 21 derives a quantization parameter corresponding to a quantization width “10” equal to or larger than the threshold value for a pixel having a pixel value “50” or more.

FIG. 3 is a diagram showing an example of a predicted image. In the following, the predicted image is composed of "8 × 8" pixel groups as an example. The derivation unit 21 quantizes the pixel values of the pixels less than the pixel value “50” of the statistical image with the quantization width “1”. The derivation unit 21 quantizes the pixel values of the pixels of the statistical image having a pixel value of “50” or more and less than “150” with a quantization width of “5”. The derivation unit 21 quantizes the pixel values of pixels having a pixel value of “150” or more in the statistical image with a quantization width of “10”.

As a result of quantization with a quantization width of "10", the pixel value of the pixel 100 of the predicted image becomes "150". As a result of the quantization with the quantization width "1", each pixel value of the pixels 101-108 of the predicted image remains "6".

FIG. 4 is a diagram showing an example of a difference image. In the following, the difference image is composed of "8 × 8" pixel groups as an example. The difference image shown in FIG. 4 represents the result (error value) obtained by subtracting the pixel value of the predicted image from the pixel value of the statistical image. The pixel value of the pixel 100 of the difference image is "6 (= 156-150)". Each pixel value of pixels 101-108 of the predicted image is "0 (= 6-6)".

FIG. 5 is a diagram (histogram) showing an example of the pixel value (error value) of the pixel 100 of the difference image. The horizontal axis shows the pixel value (error value) of each pixel quantized with the quantization width “10” in the difference image. The vertical axis shows the number of pixels quantized with a quantization width of "10" in the difference image for each pixel value (error value).

The difference quantization unit 23 detects the pixel 100 with the pixel value “6”, which is equal to or less than the quantization width “10” in absolute value, in the difference image. The difference quantization unit 23 quantizes the pixel value “6” of the pixel 100 detected in the difference image to 0.

FIG. 6 is a diagram (histogram) showing an example of each pixel value (each error value) of pixels 101-108 of the difference image. The horizontal axis shows the pixel value (error value) of each pixel quantized with the quantization width “1” in the difference image. The vertical axis indicates the number of pixels quantized with the quantization width “1” in the difference image for each pixel value (error value).

The difference quantization unit 23 detects in the difference image the pixel 100 having a pixel value “0” which is equal to or less than the quantization width “1” in comparison with an absolute value. The difference quantization unit 23 quantizes each pixel value “6” of the pixels 101-108 detected in the difference image to 0.

FIG. 7 is a diagram showing an example of a differentially decoded image. The pixel value "6" of the pixel 100 of the difference image is quantized to 0 by the difference quantization unit 23, so that the pixel value of the pixel 100 in the difference data becomes 0. Since each pixel value "6" of the pixel 101-108 of the difference image is quantized to 0 by the difference quantization unit 23, each pixel value of the pixel 101-108 in the difference data becomes 0. The difference decoding unit 31 generates the difference decoding image shown in FIG. 7 by executing the decoding process on the difference data.

FIG. 8 is a diagram showing an example of a decoded image. The addition unit 32 generates the decoded image (compressed statistical image) shown in FIG. 8 by adding the predicted image shown in FIG. 3 and the differential decoded image shown in FIG. 7 for each pixel. To do. In the decoded image shown in FIG. 8, since each pixel value of pixels 101-108 is not “0”, the statistical property of the depopulated area is maintained. Further, in the decoded image, the ratio of the statistical data (pixel values) represented by the pixels 100 and the pixels 101-108 is almost the same as the ratio of the statistical data represented by the pixels 100 and the pixels 101-108 in the statistical image. , Statistical properties are maintained in the decoded image.

Next, an operation example of the quantization device 1a will be described.
FIG. 9 is a flowchart showing an operation example of the quantization device 1a. The imaging unit 20 acquires statistical data for each position in the real space. The imaging unit 20 generates an uncompressed statistical image based on the statistical data. The imaging unit 20 outputs an uncompressed statistical image to the derivation unit 21 and the subtraction unit 22 (step S101).

The derivation unit 21 acquires an uncompressed statistical image. The derivation unit 21 derives a quantization parameter corresponding to the quantization width according to the pixel value for each pixel of the coordinate i in the part or the whole of the uncompressed statistical image. The derivation unit 21 outputs the quantization parameter to the difference quantization unit 23 for each pixel (step S102).

Deriving unit 21, the pixel value V _i of the coordinate i of the statistical image, the quantization width d _i determined according to the quantization parameter determined based on the pixel values for each pixel, to perform the quantization process .. The derivation unit 21 outputs the feature amount data obtained as a result of the quantization processing to the inverse mapping processing unit 30 (step S103).

The subtraction unit 22 acquires an uncompressed statistical image from the imaging unit 20. The subtraction unit 22 acquires the predicted image obtained as a result of the inverse mapping process from the derivation unit 21. The subtraction unit 22 subtracts the pixel value of the predicted image from the pixel value of the uncompressed statistical image for each pixel. The subtraction unit 22 outputs the subtraction result as a difference image to the difference quantization unit 23 (step S104).

The difference quantization unit 23 acquires a difference image from the subtraction unit 22. The difference quantization unit 23 acquires the quantization parameter from the derivation unit 21. Difference quantization unit 23, the absolute value of the pixel values of the difference image (error value) is the equal to or less than the absolute value of the pixel of the quantization width "± d _i" is detected in the difference image. The difference quantization unit quantizes the error value of the pixel detected in the difference image to 0. The difference quantization unit 23 outputs the difference data to the difference decoding unit 31 (step S105).

The inverse mapping processing unit 30 acquires feature data from the derivation unit 21. The inverse mapping processing unit 30 generates a predicted image based on the feature amount data. The inverse mapping processing unit 30 outputs the predicted image obtained as a result of the inverse mapping processing to the adding unit 32 (step S106).

The difference decoding unit 31 acquires the difference data from the difference quantization unit 23. The difference decoding unit 31 executes decoding processing such as inverse quantization on the difference data. The difference decoding unit 31 outputs the difference decoding image to the addition unit 32 (step S107).

The addition unit 32 acquires the feature amount data from the inverse mapping processing unit 30. The addition unit 32 acquires the difference decoding image from the difference decoding unit 31. The addition unit 32 adds the pixel value of the feature amount data and the pixel value of the difference-decoded image for each pixel. The addition unit 32 outputs the addition result to a predetermined external device (step S108).

As described above, the quantization device 1a of the first embodiment includes an imaging unit 20 and a derivation unit 21. The imaging unit 20 converts the statistical data at the position in the real space into the pixel value of the coordinates associated with the position in the statistical image. The derivation unit 21 derives the quantization parameter corresponding to the quantization width of the pixel value for each position (pixel) in the spatial region with respect to a part or the whole of the statistical image.

This makes it possible to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties (features).

Deriving unit 21 may derive a quantization parameter corresponding to the longer quantization width is larger pixel value V _i of the coordinate i 'd _{i =} f (V _i). " The difference quantization unit 23 quantizes the error value, which is the difference between the pixel value not quantized by the quantization width and the pixel value quantized by the quantization width in the difference image. Difference quantization unit 23, the absolute value of the pixel values of the difference image (error values) may detect a pixel is less than the absolute value of the quantization width "± d _i". The difference quantization unit may quantize the error value of the detected pixel to 0.

(Second Embodiment)
In the second embodiment, that the pixel value V _i at determined quantization width d _i according to the quantization parameter determined for each plurality of pixels (area) is quantized, different from the first embodiment. In the second embodiment, the differences from the first embodiment will be described.

Deriving section 21, a quantization parameter corresponding to the quantization width d _i of the pixel values of the uncompressed statistical image is derived for each pixel of a plurality of coordinate i in part or all of the statistical image. Deriving unit 21, the pixel value V _i of the plurality of coordinate i of the statistical image, executes the quantization processing by the quantization width d _i determined according to the quantization parameter determined for each plurality of pixels. Quantization width _{d i} is expressed by the equation (2).

_{_{d i = f (A i)}} ... (2)

Here, i is an index and represents the coordinates in the statistical image corresponding to the position in the real space. A _i represents the average value of the pixel values V _i of a plurality of coordinates i. f represents a predetermined proportional function. The proportional function "f" is represented by the function exemplified in the equation (3).
_{f (V i) = a ×} V i + b ... (3)

Here, a represents a predetermined coefficient. b represents a predetermined constant.

The derivation unit 21 outputs the quantization parameters defined for each region including a plurality of pixels to the difference quantization unit 23. Deriving unit 21, the quantization width d _i instead of the quantization parameter may be outputted to the difference quantization unit 23.

The difference quantization unit 23 (compression processing unit) acquires a difference image from the subtraction unit 22. The difference quantization unit 23 acquires the quantization parameters defined for each region including a plurality of pixels from the derivation unit 21. Difference quantization unit 23, the pixel values of the difference image (error value) is quantized by the quantization width d _i determined according to the quantization parameter determined for each region including a plurality of pixels.

As described above, the quantization device 1a of the second embodiment includes an imaging unit 20 and a derivation unit 21. The imaging unit 20 converts the statistical data at the position in the real space into the pixel value of the coordinates associated with the position in the statistical image. The derivation unit 21 derives the quantization parameter corresponding to the quantization width of the pixel value for each position (each region) of a plurality of spatial regions for a part or the whole of the statistical image.

This makes it possible to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties in units of multiple pixels (areas).

(Third Embodiment)
The third embodiment is different from the first embodiment and the second embodiment in that the quantization parameter corresponding to the quantization width according to the operation mode (objective information) is determined. In the third embodiment, the differences from the first embodiment and the second embodiment will be described.

FIG. 10 is a diagram showing a configuration example of the quantization device 1b. The quantization device 1b is a device that quantizes the pixel value of each pixel of the statistical image. The quantization device 1b includes a coding device 2b and a decoding device 3. The coding device 2b includes an imaging unit 20, a derivation unit 21, a subtraction unit 22, a difference quantization unit 23, and an acquisition unit 24. The decoding device 3 includes an inverse mapping processing unit 30, a difference decoding unit 31, and an addition unit 32.

The acquisition unit 24 (purpose acquisition unit) is, for example, an input device such as a keyboard or a touch panel. The acquisition unit 24 acquires the operation mode and the coordinate data. The quantization parameters are determined according to the operation mode. The operation mode is predetermined by the user according to the purpose of statistical processing. The purpose of statistical processing is, for example, to obtain statistical data on demand for determining which of a plurality of positions (regions) in real space to dispatch more taxis.

For example, in urban areas at night, it is expected that the population of residential areas, the population of office areas, and the population of downtown areas will all be high. Even in areas with similarly high populations, as people return to residential areas on weekday nights, the population of residential areas will increase, and the population of both office areas and downtown areas will decrease. That is, a flow of people occurs starting from the office district and the downtown area. Since it is desirable that many taxis are dispatched at the starting point of the flow of people, it is desirable that the populations of the office district and the downtown area be derived with higher accuracy. The value of the quantization parameter of each pixel corresponding to the office district and the downtown area is the quantization of each pixel corresponding to the residential area and the depopulated area so that each population of the office district and the downtown area can be derived with higher accuracy. It is set to a value smaller than the value of the parameter.

The quantization error allowed for the pixel values of the coordinates associated with the positions where statistical data needs to be obtained is allowed for the pixel values of the coordinates associated with the positions where statistical data does not need to be obtained. It is smaller than the quantization error. Therefore, the pixel value of the coordinates associated with the position where statistical data needs to be obtained is quantized with a quantization width smaller than a predetermined threshold value.

The coordinate data represents the coordinates of one or more pixels corresponding to one or more positions where statistical data needs to be obtained in the statistical image. The coordinate data may represent the coordinates of one or more pixels corresponding to one or more positions where statistical data does not need to be obtained in the statistical image. The acquisition unit 24 outputs the operation mode and the coordinate data to the derivation unit 21.

The derivation unit 21 acquires the operation mode and the coordinate data. The derivation unit 21 acquires an uncompressed statistical image from the imaging unit 20. Deriving section 21, a quantization parameter corresponding to the quantization width d _i of the pixel values of the uncompressed statistical image is derived for each pixel of the coordinate i indicated by the coordinate data.

When the coordinate data indicates the coordinates associated with the position where the statistical data needs to be obtained, the derivation unit 21 quantizes a value less than the threshold value for each pixel of one or more coordinates i indicated by the coordinate data. Derive the parameters. When the coordinate data indicates the coordinates associated with the positions where it is not necessary to obtain the statistical data, the derivation unit 21 quantizes the value equal to or more than the threshold value for each pixel of one or more coordinates i indicated by the coordinate data. Derive the parameters.

The quantization width corresponding to the quantization parameter having a value below the threshold value is shorter than the quantization width corresponding to the quantization parameter having a value above the threshold value. Deriving unit 21, the pixel value V _i of the coordinate i of the statistical image, executes the quantization processing by the quantization width d _i determined according to the quantization parameter determined for each pixel.

Next, an example of quantization processing will be described.
FIG. 11 is a diagram showing an example of a statistical image. In FIG. 11, the statistical image includes pixels 200-203. Pixels 200-203 are pixels with coordinates associated with each position (each region) for which statistical data needs to be obtained.

The pixel value of pixel 200 represents, for example, the number of taxi demands in the third region. The pixel value of the pixel 200 at the coordinates (3, 5) is "23" as an example. The pixel value of pixel 201 represents, for example, the number of taxi demands in the fourth region. The pixel value of the pixel 201 of the coordinates (4,5) is "25" as an example. The pixel value of pixel 202 represents, for example, the number of taxi demands in the fifth region. The pixel value of the pixel 202 at the coordinates (3, 6) is "100" as an example. The pixel value of pixel 203 represents, for example, the number of taxi demands in the sixth region. The pixel value of the pixel 203 of the coordinates (4, 6) is "102" as an example.

The permissible quantization error for each pixel value of pixels 200-203 is smaller than the permissible quantization error for the pixel values of pixels other than pixels 200-203 in the statistical image. If, for example, the same quantization width "10" is used for the quantization of the pixel values for all the pixels in the statistical image, the ratio (denseness relationship) between the statistical data (values) represented by each pixel value is Statistical properties may not be maintained in statistical images due to changes.

In order to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties in each region where statistical data needs to be obtained, the derivation unit 21 is less than the threshold value for each pixel value of the pixels 200-203. The quantization parameter corresponding to the quantization width (for example, 2) of is derived. The derivation unit 21 may derive a quantization parameter corresponding to a quantization width (for example, 10) equal to or larger than the threshold value for the pixel value of each pixel other than the pixels 200-203.

FIG. 12 is a diagram showing an example of a predicted image. The derivation unit 21 quantizes the pixel values of the pixels 200 to 203 with a quantization width “2” that is less than the threshold value, for example. The derivation unit 21 quantizes the pixel values of the pixels other than the pixels 200-203 with a quantization width “10” equal to or larger than the threshold value, for example. As a result of the quantization with the quantization width "2", the pixel value of the pixel 200 of the predicted image becomes "22". The pixel value of the pixel 201 of the predicted image is "24".

FIG. 13 is a diagram showing an example of a difference image. The difference image shown in FIG. 13 represents the result (error value) obtained by subtracting the pixel value of the predicted image from the pixel value of the statistical image. The pixel value of the pixel 200 of the difference image is "1 (= 23-22)". The pixel value of the pixel 201 of the difference image is "1 (= 25-24)". The pixel value of the pixel 202 of the difference image is "0 (= 100-100)". The pixel value of the pixel 203 of the difference image is "0 (= 102-102)".

FIG. 14 is a diagram (histogram) showing an example of each pixel value of a predetermined pixel group (pixels 200-203) of the difference image. The horizontal axis represents the pixel value (error value) of the pixels 200-203 quantized with the quantization width “2” in the difference image. The vertical axis shows the number of pixels 200-203 quantized with a quantization width of "2" in the difference image for each pixel value (error value).

The difference quantization unit 23 detects in the difference image a pixel having a quantization width of “2” or less as compared with an absolute value among the pixel values of the pixels 200 to 203 indicated by the coordinate data in the difference image. To do. The difference quantization unit 23 quantizes the pixel value of each pixel detected in the difference image to 0. The difference quantization unit 23 may quantize with a quantization width of "10" so that the pixel value of each pixel other than the pixels 200-203 is set to 0 in the difference image.

FIG. 15 is a diagram showing an example of an image representing the difference data. The pixel value "1" of the pixels 200-201 of the difference image is quantized to 0 by the difference quantization unit 23, so that the pixel value of the pixel 200 in the difference data becomes 0. The difference decoding unit 31 generates the difference decoding image shown in FIG. 15 by executing the decoding process on the difference data.

FIG. 16 is a diagram showing an example of a decoded image. The decoded image (compressed statistical image) shown in FIG. 16 is generated by adding the predicted image shown in FIG. 12 and the differential decoded image shown in FIG. 15 for each pixel. Since the ratio of the statistical data (pixel values) represented by the pixels 200-203 in the decoded image shown in FIG. 16 is almost the same as the ratio of the statistical data represented by the pixels 200-203 in the statistical image, the statistics in the decoded image Characteristic properties are maintained.

Next, an operation example of the quantization device 1b will be described.
FIG. 17 is a flowchart showing an operation example of the quantization device 1b. Step S201 shown in FIG. 17 is the same operation as step S101 shown in FIG. The acquisition unit 24 acquires the operation mode and the coordinate data. The acquisition unit 24 outputs the operation mode and coordinate data to the derivation unit 21 (step S202).

The derivation unit 21 acquires the operation mode and coordinate data. The derivation unit 21 derives the quantization parameter corresponding to the quantization width according to the operation mode for each pixel indicated by the coordinate data. The derivation unit 21 outputs the quantization parameter to the difference quantization unit 23 for each pixel (step S203). Step S201 shown in FIG. 17 is the same operation as step S101 shown in FIG.

As described above, the quantization device 1b of the third embodiment further includes an imaging unit 20, a derivation unit 21, and an acquisition unit 24. The acquisition unit 24 acquires information (for example, an operation mode) representing the purpose of statistical processing. The derivation unit 21 derives the quantization parameter based on the position in the real space and the information representing the purpose of the statistical processing. ..

This makes it possible to quantize the pixel values of the pixels of the statistical image so as to maintain the statistical properties according to the operation mode (purpose of statistical processing).

For example, even if detailed statistical data is required only for a given area regardless of the absolute number of population in each area, the pixel values of the pixels of the statistical image are set so as to maintain the statistical properties for the given area. It is possible to quantize.

For example, statistical processing to control taxi dispatch may require more detailed statistical data for areas with a large absolute number of passenger candidates, but statistical properties for areas with a large absolute number of passenger candidates. It is possible to quantize the pixel values of the pixels of the statistical image so as to keep the above.

(Modification example)
Each function f shown in the equation (1), the equation (2) or the equation (3) may be a function other than the proportional function. When there are multiple reference values (centers of the distribution range of pixel values) of a plurality of quantized pixel values in a statistical image, the quantization width is determined for each reference value of the plurality of quantized pixel values. May be good. Multiple pixel values quantized in the statistical image according to the distribution range of the plurality of pixel values quantized in the statistical image and the distribution range of the plurality of pixel values (error values) quantized in the difference image. The reference value of may be determined. The function "f" may be, for example, the function shown in the equation (4) expressed using the logarithm "log".

_{f (V i) = a ×} log (V i) ... (4)

Compared with the functions of Eqs. (1), (2) and (3), the function of Eq. (4) can increase the change in the quantization width with the change of the value. That is, the function of Eq. (4) can maintain the accuracy of extremely large values. The function of equation (4) may be expressed using, for example, a power instead of being expressed using logarithms.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs and the like within a range that does not deviate from the gist of the present invention.

The present invention is applicable to an image processing apparatus that performs image coding and image decoding.

1a, 1b ... Quantizer, 2a, 2b ... Encoding device, 20 ... Imaging unit, 21 ... Derivation unit, 22 ... Subtraction unit, 23 ... Difference quantization unit, 24 ... Acquisition unit, 30 ... Inverse mapping processing unit , 31 ... difference decoding unit, 32 ... addition unit, 100 to 108 ... pixels, 200 to 203 ... pixels

Claims

An imaging unit that converts statistical data at a position in real space into pixel values of coordinates associated with the position in an image.
A quantization device including a derivation unit that derives a quantization parameter corresponding to the quantization width of the pixel value for each one or more positions in the real space for a part or the whole of the image.
Further equipped with a purpose acquisition unit for acquiring information indicating the purpose of statistical processing.
The quantization device according to claim 1, wherein the derivation unit derives the quantization parameter based on the position in the real space and the information representing the purpose of the statistical processing.
The quantization device according to claim 1 or 2, wherein the derivation unit derives the quantization parameter corresponding to the quantization width, which is longer as the pixel value is larger.
Further provided with a difference quantization unit that quantizes an error value that is a difference between the pixel value that is not quantized by the quantization width and the pixel value that is quantized by the quantization width.
The difference quantization unit detects a pixel whose absolute value of the error value is equal to or less than the absolute value of the quantization width, and quantizes the error value of the detected pixel to 0, according to claim 1. Item 3. The quantization apparatus according to any one of items 3.
It is a quantization method executed by a quantization device.
A step of converting statistical data at a position in real space into a pixel value of coordinates associated with the position in an image.
A quantization method including a step of deriving a quantization parameter corresponding to the quantization width of the pixel value for each one or more positions in the real space for a part or the whole of the image.
A program for operating a computer as the quantization device according to any one of claims 1 to 4.